Introduction

Ecological communities are the result of species’ geographic distribution, dispersal capabilities, speciation, environmental filtering, and stochastic events. By definition stochasticic events are random colonization and extinction events that ultimately cause ecological drift. Deterministic processes are non-random environmental processes like environmental filtering, biotic interactions, and niche partitioning. Wetlands isolated in uplands may act as islands, so predictions about stochasticity made by the Theory of Island Biogeography may apply. The following predictions will be tested throughout the Eastern Highland Rim ecoregion:

  1. Wetland size will positively correlate with species richness.

  2. The distance between wetlands will negatively correlate with the similarity of their plant communities.

  3. Differences in plant communities will not significantly differ from a randomly-generated null model.


Methods

Data Collection

Plots were sampled at 27 independent sites throughout a subset of the Eastern Highland Rim ecoregion (delineated in yellow below). Note that some sites were in close proximity, sometimes as close together as 200m apart.

Data was collected for plots using 1 m^2 quadrats, of which there were 1031 total. With a few exceptions, all vascular plant species within each quadrat were identified to the species level. Variables were measured at each site; these include water permanence, early season water depth, late season water depth, water depth change, site area, and land use (grazing).



Data Configuration

Nearly all statistical packages require the data to be in a presence-absence form, typically as matrix with Site as a column so subsamples can be organized accordingly. Rows are named using quadrat codes (plot, transect, quadrat).

Table 1. Sample of data in presence-absence format.
Site CXJOOR JUNDEB JUNREP PROPEC SCICYP CXLOUI
1.1.1 1 1 1 1 1 1 0
1.1.2 1 0 1 0 1 1 1
1.1.3 1 0 0 0 1 1 1
1.1.4 1 1 0 0 1 1 1
1.1.5 1 0 0 0 0 0 1

R Libraries Used

Analyses
-iNEXT, vegan, mclust, MuMIn, lme4

Display/Data Wrangling
-ggplot2, dplyr, tidyr, reshape, knitr, kableExtra, ggfortify, ggpubr, colordistance, otuSummary, stringi, stringr, ggextra, gridextra

Mapping
-geosphere, rgdal, ggmap

Awesomeness
-beepr, wesanderson


Richness

Sites had widely varying observed total richness.

Extrapolation of Richness

The use of extrapolated species accumulation curves can tell us how many species are likely to be at the site based on how many were found in accumulating subsamples.

These curves illustrate not only where the flattening point (expected richness) occurs, but also how quickly. Examining a curve can allow someone to estimate how many more samples would be needed to reach that point, however if doing so samples a larger area then the curve may never flatten.

Sampling Effort

Sampling effort (# quadrats/area sampled) did not affect percent estimated sampling completion.

There is no relationship between sampling effort and completion percentage (F= 0.335, p=0.568).

Model Selection

A negative binomial generalized linear models was used because the response variable (richness) is count data and cannot be negative. Models were made to determine which combinations of variables best explained differences in site richness.

Comparison of models.
Model K df logLik AICc delta
model8 Perm * StrtDpth * DpthChng na.fail 8 -124.21 272.43 0.000
model7 Perm * StrtDpth * DpthChng + Grzng 9 -123.08 274.75 2.323
model6 Perm * StrtDpth * DpthChng + Area 9 -123.24 275.07 2.642
model1 Perm * StrtDpth * DpthChng + Grzng + Area 10 -121.96 277.66 5.235
model9 Perm * StrtDpth + StrtDpth * DpthChnge + DpthChng * Perm 7 -135.93 291.75 19.319
model4 StrtDpth * DpthChng + Grzng + Area 6 -139.04 294.28 21.852
model2 Perm * DpthChng + Grzng + Area 6 -159.06 334.31 61.882
model3 Perm * StrtDpth + Grzng + Area 6 -159.87 335.95 63.519
model5 Perm + StrtDpth + DpthChng + Grzng + Area 6 -166.93 350.05 77.625


The fully crossed model of permanence, early season water depth, and depth change had the most explanatory power, but it contains many interaction terms and should be trimmed down if possible.


Model 8 was dredged, but the markdown failed to knit when this was included. A screenshot of the resulting table was included instead. Variables are abbreviated as follows:

dc= depth change p= water permanence sd= early season water depth



The fully crossed model of permanence, early season water depth, and depth change still had the most explanatory power. This includes a three way interaction, which will make interpretation difficult but it is biologically relevant so it should be included.


Beta Diversity


Beta diversity is a measure of compositional similarity ranging from 0 (sites have the same composition) to 1 (no species overlap between sites). There are multiple ways to measure beta diversity; I used Sorensen’s Dissimilarity Index because I had binary presence-absence data. Beta diversity is calculated as pairwise comparisons, meaning that the output is in matrix-form. This makes it difficult to interpret as-is. Wes Anderson’s sweet colors can help, however.

Geographic Distance

One popular prediction of the Theory of Island Biogeography (and others) is that it is easier for individuals to disperse shorter distances and this results in close sites being more similar than ones far apart. We will test this using a Mantel test, which will test the correlation between the beta diversity matrix above and a geographic distance matrix. I had the test use Spearman’s rank correlation, which is the nonparametric version of the test because the data aren’t normal. This test found that geographic distance and species similarity are not correlated (rm= 0.753,p= 0.139).

Site Variables

A PerMANOVA (Permutational Multivariate Analysis Of Variance) is a nonparametric test that can be used to detect whether the measured variables significantly influence beta diversity. The syntax of this is similar to a GLM, though instead of a response variable you use a response matrix.

Effect of variables on beta diversity. B=9999. The sum r^2 of the variables accounted for by the model is 0.459.
Df SumsOfSqs MeanSqs F.Model R2 Pr(>F)
Permanence 1 0.535 0.535 1.879 0.060 0.019
StartDepth 1 0.371 0.371 1.302 0.041 0.145
DepthChange 1 0.763 0.763 2.681 0.085 0.001
Grazing 1 0.671 0.671 2.357 0.075 0.003
Area 1 0.424 0.424 1.489 0.047 0.071
Permanence:StartDepth 1 0.281 0.281 0.986 0.031 0.466
Permanence:DepthChange 1 0.310 0.310 1.089 0.035 0.337
StartDepth:DepthChange 1 0.427 0.427 1.500 0.048 0.068
Permanence:StartDepth:DepthChange 1 0.333 0.333 1.170 0.037 0.245
Residuals 17 4.841 0.285 NA 0.541 NA
Total 26 8.956 NA NA 1.000 NA


It is difficult to visualize the trends from the PerMANOVA results alone, so they were each plotted by variable. An OLS regression line was added to continuous variables to get a feel for the general trend, but remember: these are not linear relationships!

The coloration of the points corresponds to their corresponding Raup-Crick value. This value is the permutated probability that the beta diversity was the result of deterministic (non-random) processes.


Takeaways

  1. Area was a poor predictor of richness after accounting for other variables. Hydrological variables were much better predictors of richness.

  2. Distance between sites did not significantly correlate with community similarity between sites.

  3. Community similarity was significantly affected by water permanence, water depth change, and grazing.

  4. Stochasticity is dependent on site conditions. Sites that are small, shallow, grazed, and/or experience large seasonal changes in water depth express a larger stochastic signal.